-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathimage_classification_model.py
249 lines (199 loc) · 11.1 KB
/
image_classification_model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
import torch.nn as nn
from torchvision import models
from torchvision.models import EfficientNet_B0_Weights, EfficientNet_B1_Weights, EfficientNet_B2_Weights, \
EfficientNet_B3_Weights, EfficientNet_B4_Weights, EfficientNet_B5_Weights, EfficientNet_B6_Weights, \
EfficientNet_B7_Weights, ResNet18_Weights, ResNet34_Weights, ResNet50_Weights, ResNet101_Weights, ResNet152_Weights, \
EfficientNet_V2_M_Weights, EfficientNet_V2_S_Weights, EfficientNet_V2_L_Weights
import os
import torch
from src import cct_14_7x2_384, cct_14_7x2_224, cct_7_7x2_224_sine
from torchinfo import summary
'''
Reference for Compact Convolutional Transformers
https://github.com/SHI-Labs/Compact-Transformers
'''
class ImageClassificationModel(nn.Module):
def __init__(self, cfg):
super().__init__()
self.num_classes = cfg["model"]["num_classes"]
self.n_nodes = cfg["model"]["n_nodes"]
self.dropout = cfg["model"]["dropout"]
self.freeze_layers = cfg["model"].get("freeze_layers", 1.0) > 0.0
self.image_size = cfg["data"]["size"]
self.pretrained = cfg["model"].get("pretrained", 1.0) > 0.0
self.model = self.get_model(cfg["model"]["name_pretrained_model"])
print("")
summary(model=self.model,
input_size=(1, 3, self.image_size, self.image_size),
col_names=["input_size", "output_size", "num_params", "trainable"],
col_width=20,
row_settings=["var_names"])
print("")
# connetto i vari strati definendo la funzione forward
def forward(self, x):
x = self.model(x)
return x
def classifier_head(self, output_dim):
return nn.Sequential(nn.Flatten(),
nn.Linear(output_dim, self.n_nodes),
nn.ReLU(),
nn.Dropout(self.dropout),
nn.Linear(self.n_nodes, self.num_classes))
def freeze_layers_base_model(self, model):
for name, param in model.named_parameters():
param.requires_grad = False
def get_model(self, name_pretrained_model):
if name_pretrained_model == 'resnet18':
base_model = models.resnet18(weights=ResNet18_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.fc = self.classifier_head(512)
elif name_pretrained_model == 'resnet34':
base_model = models.resnet34(weights=ResNet34_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.fc = self.classifier_head(512)
elif name_pretrained_model == 'resnet50':
base_model = models.resnet50(weights=ResNet50_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.fc = self.classifier_head(2048)
elif name_pretrained_model == 'resnet101':
base_model = models.resnet101(weights=ResNet101_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.fc = self.classifier_head(2048)
elif name_pretrained_model == 'resnet152':
base_model = models.resnet152(weights=ResNet152_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.fc = self.classifier_head(2048)
elif name_pretrained_model == 'efficientnet_b0':
base_model = models.efficientnet_b0(weights=EfficientNet_B0_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(1280)
elif name_pretrained_model == 'efficientnet_b1':
base_model = models.efficientnet_b1(weights=EfficientNet_B1_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(1280)
elif name_pretrained_model == 'efficientnet_b2':
base_model = models.efficientnet_b2(weights=EfficientNet_B2_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(1408)
elif name_pretrained_model == 'efficientnet_b3':
base_model = models.efficientnet_b3(weights=EfficientNet_B3_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(1536)
elif name_pretrained_model == 'efficientnet_b4':
base_model = models.efficientnet_b4(weights=EfficientNet_B4_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(1792)
elif name_pretrained_model == 'efficientnet_b5':
base_model = models.efficientnet_b5(weights=EfficientNet_B5_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(2048)
elif name_pretrained_model == 'efficientnet_b6':
base_model = models.efficientnet_b6(weights=EfficientNet_B6_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(2304)
elif name_pretrained_model == 'efficientnet_b7':
base_model = models.efficientnet_b7(weights=EfficientNet_B7_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(2560)
elif name_pretrained_model == 'efficientnet_v2_m':
base_model = models.efficientnet_v2_m(weights=EfficientNet_V2_M_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(1280)
elif name_pretrained_model == 'efficientnet_v2_s':
base_model = models.efficientnet_v2_s(weights=EfficientNet_V2_S_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(1280)
elif name_pretrained_model == 'efficientnet_v2_l':
base_model = models.efficientnet_v2_l(weights=EfficientNet_V2_L_Weights.DEFAULT)
base_model.avgpool = nn.AdaptiveAvgPool2d(output_size=(1, 1))
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
base_model.classifier = self.classifier_head(1280)
elif name_pretrained_model == 'cct_14_7x2_224':
base_model = cct_14_7x2_224(pretrained=self.pretrained, progress=self.pretrained, num_classes=self.num_classes, img_size=self.image_size)
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
elif name_pretrained_model == 'cct_14_7x2_384':
base_model = cct_14_7x2_384(pretrained=self.pretrained, progress=self.pretrained, num_classes=self.num_classes, img_size=self.image_size)
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
elif name_pretrained_model == 'cct_7_7x2_224_sine':
base_model = cct_7_7x2_224_sine(pretrained=self.pretrained, progress=self.pretrained, num_classes=self.num_classes, img_size=self.image_size)
if self.freeze_layers:
print("Freeze layers of pretrained model")
self.freeze_layers_base_model(base_model)
return base_model
def find_last_checkpoint_file(checkpoint_dir, use_best_checkpoint=False):
'''
Cerco nella directory checkpoint_dir il file .pth.
Se use_best_checkpoint = True prendo il best checkpoint
Se use_best_checkpoint = False prendo quello con l'epoca maggiore tra i checkpoint ordinari
:param checkpoint_dir:
:param use_best_checkpoint:
:return:
'''
print("Cerco il file .pth in checkpoint_dir {}: ".format(checkpoint_dir))
list_file_paths = []
for file in os.listdir(checkpoint_dir):
if file.endswith(".pth"):
path_file = os.path.join(checkpoint_dir, file)
list_file_paths.append(path_file)
print("Find: ", path_file)
print("Number of files .pth: {}".format(int(len(list_file_paths))))
path_checkpoint = None
if len(list_file_paths) > 0:
if use_best_checkpoint:
if os.path.isfile(os.path.join(checkpoint_dir, 'best.pth')):
path_checkpoint = os.path.join(checkpoint_dir, 'best.pth')
else:
if os.path.isfile(os.path.join(checkpoint_dir, 'latest.pth')):
path_checkpoint = os.path.join(checkpoint_dir, 'latest.pth')
return path_checkpoint